security-maintainer

Universal security maintenance skill for Windows, macOS, and Linux/WSL. Audits host and environment software, correlates with vulnerability feeds (OSV, CISA KEV, VulnCheck KEV, FIRST EPSS, NIST NVD, MSRC), audits AI & Agent security (OWASP LLM Top 10, prompt injections, unauthenticated local LLM services, MCP tool poisoning, API key exposure), and executes cryptographic-hash approved updates with post-remediation verification.

Security Maintainer for OpenClaw

Universal OpenClaw skill engineered to audit software and dependencies across Windows, macOS, Linux distributions, and WSL environments. It correlates installed packages with authoritative vulnerability feeds (OSV.dev, CISA KEV, VulnCheck KEV, FIRST EPSS, NIST NVD, MSRC, and Linux security trackers), inspects unauthenticated local AI services and agent configurations, and executes cryptographically authorized remediations with strict post-installation verification.

[!NOTE] Language & Locale Support: This skill documentation and CLI interface are standardized in English as the universal primary language. Users can invoke actions using canonical English or Spanish triggers with identical deterministic behavior.

Core Security Rules

• Strict Default ask Mode: System modifications are never applied without first generating a numbered, cryptographically hashed (SHA-256) plan.
• Bound and Expiring Approvals: Approval is strictly bound to the exact SHA-256 fingerprint of the proposed actions and expires after 30 minutes. If the host state changes, a new plan must be generated.
• Zero Silent Privilege Escalation: If an update requires elevated Administrator privileges on Windows or root rights in WSL/Linux and these are not pre-authorized, the action is marked as BLOCKED_PERMISSIONS with clear manual guidance.
• Safe Advisory Handling: Remote CVE, USN, and advisory content is treated strictly as untrusted metadata. No remote script or command extracted from feeds is ever evaluated or executed.
• Strict Finding Classification:
  • General Update Available (no active CVE).
  • Official Security Patch (resolves known CVEs/GHSA).
  • Configuration Hardening (does not modify binaries).
  • Vulnerability Without Fix (unpatched or EOL software).
  • Unsupported / End-of-Life component.
• Reboots Disabled by Default: System reboots are never performed without explicit confirmation.
• No Destructive Operations: The skill never uninstalls packages or deletes project directories.

---

Operating Modes

1. audit (Read-Only):

   • Collects host inventory across Windows (WinGet, Registry, Windows Update), macOS (Homebrew, MAS), and Linux/WSL (apt, dpkg, apk, dnf, rpm).
   • Cross-references packages against OSV.dev and KEV feeds.
   • Inspects network-listening AI services (Ollama, LM Studio) and local MCP configs when --ai-defense or --all is set.
   • Generates structured Markdown and JSON reports without modifying the environment.

2. ask (Default Guided Mode):

   • Executes audit phase and computes a deterministic action plan.
   • Generates a UUIDv4 plan ID and SHA-256 hash. Waits for explicit operator confirmation.

3. auto (Pre-Authorized Policy):

   • Applies pre-approved actions only if explicitly configured in persistent policy (default_config.yaml).

---

Primary CLI Commands

bash

# 1. Complete read-only audit
python3 -m security_maintainer.cli audit --all

# 2. Generate remediation plan (ask mode)
python3 -m security_maintainer.cli plan --environment all

# 3. Apply approved plan with verified fingerprint
python3 -m security_maintainer.cli apply --plan-id <ID> --plan-hash <HASH>

# 4. Verify post-remediation effective versions
python3 -m security_maintainer.cli verify --environment all --plan-id <ID>

# 5. Display latest report
python3 -m security_maintainer.cli report

---

OpenClaw Invocation Workflow

Users invoke the skill using explicit canonical triggers:

1. security-maintainer: audit system software and vulnerabilities:
   • Executes read-only audit and presents structured coverage and vulnerability findings.
2. security-maintainer: audit ai agent security and local llm services:
   • Inspects AI configuration files, prompts for injection patterns, and verifies unauthenticated local LLM endpoints.
3. security-maintainer: generate security remediation plan:
   • Constructs a numbered plan, prints the SHA-256 fingerprint, and waits for user approval.
4. security-maintainer: apply approved security plan:
   • Requires --plan-id and --plan-hash to verify cryptographic integrity before executing updates.
5. security-maintainer: verify status of security patches:
   • Inspects effective installed package versions and validates that vulnerabilities are resolved.